Related Experiment Video
Updated: May 2, 2026
08:40
Quantitation of Protein Expression and Co-localization Using Multiplexed Immuno-histochemical Staining and Multispectral Imaging
Published on: April 8, 2016
12.9K
MERGE: Multi-faceted Hierarchical Graph-based GNN for Gene Expression Prediction from Whole Slide Histopathology
Aniruddha Ganguly1, Debolina Chatterjee2, Wentao Huang1
1Stony Brook University, NY, USA.
Proceedings. IEEE Computer Society Conference on Computer Vision and Pattern Recognition
|August 28, 2025
Summary
MERGE, a novel graph neural network method, enhances gene expression prediction from whole slide images by modeling tissue location interactions. It improves accuracy by considering spatial and morphological features for better predictions.
Area of Science:
- Computational Biology
- Bioinformatics
- Genomics
Background:
- Spatial Transcriptomics (ST) integrates histology images with gene expression data.
- ST enables gene expression prediction from tissue image patches.
- Existing methods do not fully utilize inter-tissue location interactions for joint prediction.
Purpose of the Study:
- To introduce MERGE (Multi-faceted hiErarchical gRaph for Gene Expressions) for improved gene expression prediction from whole slide images (WSIs).
- To leverage interactions between different tissue locations for enhanced prediction accuracy.
- To evaluate data smoothing techniques for ST data artifact mitigation.
Main Methods:
- MERGE utilizes a multi-faceted hierarchical graph construction strategy with graph neural networks (GNNs).
- Tissue image patches are clustered based on spatial and morphological features.
- Intra- and inter-cluster edges are incorporated to model interactions between distant tissue locations.
Main Results:
- MERGE outperforms state-of-the-art techniques in gene expression prediction across multiple metrics.
- Gene-aware smoothing methods are recommended for biologically justified artifact mitigation in ST data.
- The GNN approach effectively captures interactions between spatially distant tissue regions.
Conclusions:
- MERGE provides a powerful framework for enhancing gene expression prediction from WSIs by modeling complex spatial relationships.
- The study highlights the importance of considering inter-tissue location interactions for accurate predictions.
- Gene-aware smoothing is crucial for reliable ST data analysis.
